Ranks Bhopal’s immersion ghats for re-testing against CPCB Class B thresholds, linking every number to its observation, CSV line and report page.
The problem
BMC officers need to decide which Bhopal immersion ghats to re-test first before festivals, but evidence is scattered across multiple MPPCB reports and missing data is easy to misread. We solve this by turning 2023–2025 historical monitoring into a transparent, citation-backed priority view, so officers can quickly identify higher-risk ghats, see data gaps clearly, and avoid unsupported safety or causation claims.
How it works
NeerSetu is a single workflow decision-support app for one user persona: a BMC officer. It ingests and normalizes 3 years of ghat observations, compares DO/BOD/pH against CPCB Class B screening thresholds, computes a disclosed concern index, and ranks ghats for re-testing priority. Every claim is grounded: each number links to observation ID, CSV line, and report PDF page. Missing values remain missing (never treated as zero/clean), and the app separates historical concern from sampling coverage and 2026 status. The product includes ranking, per-ghat drill-down, evidence tables/charts, and bilingual officer brief generation with strict validation guardrails.
How Claude was used
We used Claude/Claude Code during build and iteration for architecture planning, code generation/refactoring, test authoring, and documentation tightening. Claude-assisted edits were kept under deterministic validation rules, so model output never changed scoring logic or invented evidence. We also integrated optional AI narrative assistance in the app through provider adapters and strict output validation. In production flow, both Anthropic-compatible and Cerebras-compatible paths are supported; if AI fails validation or is unavailable, the system automatically falls back to deterministic template wording.
Team
NeerSetu